Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1166438 > unrolled thread

[PATCH 1/1] lib: small update for strlen, strnlen, use less cpu instructions

Back to article view | Back to linux.kernel

Contents

  [PATCH 1/1] lib: small update for strlen, strnlen, use less  cpu instructions Orestes Leal Rodriguez <lukes357@gmail.com> - 2015-06-17 00:00 +0200
    Re: [PATCH 1/1] lib: small update for strlen, strnlen, use less cpu  instructions Richard Weinberger <richard@nod.at> - 2015-06-17 22:10 +0200

#1166438 — [PATCH 1/1] lib: small update for strlen, strnlen, use less cpu instructions

 > Use the force^Wcheckpatch.pl.
This is the output of checkpatch.pl:
output of checkpatch: total: 0 errors, 0 warnings, 42 lines checked
/root/string.c.patch has no obvious style problems and is ready for 
submission

 > You need to explain that in the commit message, my young padawan.
Very small update to strlen and strnlen that now use less cpu 
instructions by using a counter to avoid memory address
arithmetic, which cause that the compiler adds more machine
instructions for computing the length of the string just before
returning from the functions, the old machine code is like the
following:

  mov    -0x4(%ebp),%edx
  mov    0x8(%ebp),%eax
  sub    %eax,%edx
  mov    %edx,%eax
  leave
  ret


now in the new versions the value is not calculated anymore,
instead  he value of the counter is put on eax after the
condition inside the loop no longer holds, and then return:

  mov    -0x4(%ebp),%eax
  leave
  ret

With this a few cpu instructions are saved.


Signed-off-by: Orestes Leal Rodriguez <lukes357@gmail.com>
---

Signed-off-by: Orestes Leal Rodriguez <lukes357@gmail.com>
diff --git a/lib/string.c b/lib/string.c
index 992bf30..c873436 100644
--- a/lib/string.c
+++ b/lib/string.c
@@ -17,6 +17,10 @@
   * * Sat Feb 09 2002, Jason Thomas <jason@topic.com.au>,
   *                    Matthew Hawkins <matt@mh.dropbear.id.au>
   * -  Kissed strtok() goodbye
+ *
+ * * Tuesday June 16 2015, Orestes Leal Rodriguez <lukes357@gmail.com>
+ * - strlen, strnlen: by using a single counter we use less cpu 
instructions
+ *   by avoiding substracting the memory addresses before return
   */

  #include <linux/types.h>
@@ -401,11 +405,11 @@ EXPORT_SYMBOL(strim);
   */
  size_t strlen(const char *s)
  {
-	const char *sc;
+	size_t sz = 0;

-	for (sc = s; *sc != '\0'; ++sc)
-		/* nothing */;
-	return sc - s;
+	for (; *s++ != '\0'; sz++)
+		/* empty */;
+	return sz;
  }
  EXPORT_SYMBOL(strlen);
  #endif
@@ -418,12 +422,13 @@ EXPORT_SYMBOL(strlen);
   */
  size_t strnlen(const char *s, size_t count)
  {
-	const char *sc;
+	size_t sz = 0;

-	for (sc = s; count-- && *sc != '\0'; ++sc)
-		/* nothing */;
-	return sc - s;
+	for (; count-- && *s++ != '\0'; sz++)
+		/* empty */;
+	return sz;
  }
+
  EXPORT_SYMBOL(strnlen);
  #endif


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [next] | [standalone]

#1167129 — Re: [PATCH 1/1] lib: small update for strlen, strnlen, use less cpu instructions

Am 17.06.2015 um 00:51 schrieb Orestes Leal Rodriguez:
>> Use the force^Wcheckpatch.pl.
> This is the output of checkpatch.pl:
> output of checkpatch: total: 0 errors, 0 warnings, 42 lines checked
> /root/string.c.patch has no obvious style problems and is ready for submission

But it does not apply at all.
Did you test it? I fear your mail client did some whitespace damage.

>> You need to explain that in the commit message, my young padawan.
> Very small update to strlen and strnlen that now use less cpu instructions by using a counter to avoid memory address
> arithmetic, which cause that the compiler adds more machine
> instructions for computing the length of the string just before
> returning from the functions, the old machine code is like the
> following:
> 
>  mov    -0x4(%ebp),%edx
>  mov    0x8(%ebp),%eax
>  sub    %eax,%edx
>  mov    %edx,%eax
>  leave
>  ret
> 
> 
> now in the new versions the value is not calculated anymore,
> instead  he value of the counter is put on eax after the
> condition inside the loop no longer holds, and then return:
> 
>  mov    -0x4(%ebp),%eax
>  leave
>  ret
> 
> With this a few cpu instructions are saved.

x86_32 does not matter here as we have already an optimized strlen() in
arch/x86/lib/string_32.c.
Did you check whether the optimization is worth on other archs?
Hint: grep __HAVE_ARCH_STRLEN

> 
> Signed-off-by: Orestes Leal Rodriguez <lukes357@gmail.com>
> ---
> 
> Signed-off-by: Orestes Leal Rodriguez <lukes357@gmail.com>

What does this 2nd SoB here?

> diff --git a/lib/string.c b/lib/string.c
> index 992bf30..c873436 100644
> --- a/lib/string.c
> +++ b/lib/string.c
> @@ -17,6 +17,10 @@
>   * * Sat Feb 09 2002, Jason Thomas <jason@topic.com.au>,
>   *                    Matthew Hawkins <matt@mh.dropbear.id.au>
>   * -  Kissed strtok() goodbye
> + *
> + * * Tuesday June 16 2015, Orestes Leal Rodriguez <lukes357@gmail.com>
> + * - strlen, strnlen: by using a single counter we use less cpu instructions
> + *   by avoiding substracting the memory addresses before return

No need to add anything here. These days we have git. :-)

>   */
> 
>  #include <linux/types.h>
> @@ -401,11 +405,11 @@ EXPORT_SYMBOL(strim);
>   */
>  size_t strlen(const char *s)
>  {
> -    const char *sc;
> +    size_t sz = 0;
> 
> -    for (sc = s; *sc != '\0'; ++sc)
> -        /* nothing */;
> -    return sc - s;
> +    for (; *s++ != '\0'; sz++)
> +        /* empty */;

<nitpick>
Why suddenly "empty" instead of "nothing"?
</nitpick>

> +    return sz;
>  }
>  EXPORT_SYMBOL(strlen);
>  #endif
> @@ -418,12 +422,13 @@ EXPORT_SYMBOL(strlen);
>   */
>  size_t strnlen(const char *s, size_t count)
>  {
> -    const char *sc;
> +    size_t sz = 0;
> 
> -    for (sc = s; count-- && *sc != '\0'; ++sc)
> -        /* nothing */;
> -    return sc - s;
> +    for (; count-- && *s++ != '\0'; sz++)
> +        /* empty */;

Same here.

Thanks,
//R2D2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web